In most sports, the physical part is paramount, and the mental part is secondary. Chess is almost entirely mental; the only physical aspect is not dozing off or dying from phlebitis before the game is over.
Hehe. Of course chess is mental, but that doesn't take away that just as you can blunder by fumbling a ball in baseball, you can blunder by fumbling a position in chess. There are hundreds of examples of (grand)masters losing a game to a much lower ranking opponent. In chess there is the added stress of not only playing against an opponent but playing against a clock also, and it's the mental stress that's mostly responsible for the mistakes you make, not only in chess but in a wide range of physical sports also.
Some? Crossword puzzles maybe. When I run into problems, I usually try to remember similar situations and the various outcomes (or at least I think I do).
That would be a concious process, but the crosswords you mention is perhaps (interestingly) related to where I mostly detect these unconcious processes in myself, namely when I'm trying to remember words (in a foreign language) and catch myself mumbling things like: malag maleg malig ahh, it's malignant. I'm unconciously doing a brute force on the phonemes of a word until it "sounds" (feels) right. There are other examples, but I've seen other people do this too, so maybe you're familliar with it also.
The rules of warfare are myriad and changeable, and the possible environments nearly infinite. The rules of chess are fixed and the battlefield unchangeable.
Agreed, but we've already discussed the number of possible positions in a normal game of chess. Isn't it fair to call these a myrad of possibilities also?;)
But all that still does not address your claim that anything that reduces a human's cognitive workload is capable of thinking.
Let's simply define thinking as excercizing your intelligence(s), IOW making use of your abilities to abstract, solve problems, etc. Any machine that reduces the need for a human to perform these tasks while the abstract models are still being produced and/or the problems are still being solved, must be performing (part of) these tasks for the human; because performing these tasks is defined as thinking, the machine must be thinking to a certain extent.
Well, I'm sure we've broken my previous record for ongoing arguments on Slashdot, and it seems we've agreed to disagree. I'd wish you a happy Thanksgiving, but your spelling of "labour" suggests you're a Brit, and the holiday is only for us colonists.:)
I'm much better at starting discussions than at ending them, and even though I tried to keep this last post short, look what happened. I'm not a Brit but a Dutchie by the way; it was them Brits who gave us trouble in Nieuw Amsterdam and rebaptized it to New York, so don't get me started about them:) Pleasant holidays.
Mallory can configure her LDAP server to not only give her root on you box, but also to remotely mount filesystems on your box. Mallory mounts her trojans over your bin directory and waits for you to start one, or also mounts a root crontab that starts a trojan automaticly. No your box IS running services, and Mallory owns it.
As long as computers are only just as good as the best human, no human could win. You are correct, as I've stated from the beginning, I don't see the relevance.
Ok, I'll try to explain it one more time. In chess (like in most sports) there is absolutely no guarantee that the better player (the higher ranking player) will always win. Even worse, you can't even prove that looking ahead further in the search tree produces better moves (because of horizon effects, the move you think is good might prove very bad if you look ahead one move further, ad nauseam). The only way of being absolutely unbeatable is to play perfect chess, IOW to exhaust the search tree completely.
I would think that someone who believes computers can think would see the waste in brute force solutions and be able to imagine the future of expert systems.
I don't see it as waste, in fact I think some of the thinking we humans do is of brute-force type, we just don't normally realize it because they are subconcious processes. The problem with rule-based systems is emergent properties; if you combine some individually simple rules, complex situations can arrise. That's why expert systems don't work for problems like chess: it's like feeding a general with knowledge of famous battles of the past and expecting him to win every future battle.
In earlier years, I built a calculator and a computer from scratch using wire-wrap, so I can attest that a calculator is nothing more than an electronic abacus or slide rule, and a computer is nothing more than a fancy calculator.
Yes, but very similarly, if you disect/study a living human brain you will only find braincells but no thought. If a brain is only neurons, how can it think? My answer: intelligence is an emergent property (also).
As a fair but untalented chess player, I can say not many.
And as a chess player that really played competition and not just recreatively, I can say a lot. There are a lot of people who practice chess every day and study it two hours a day or more.
I can understand your point (though I don't agree completely) that it's not a lot of fun to recreatively play chess against a much stronger opponent. But if you play chess competatively and want to improve your tactics, there is nothing like playing a lot of games against a strong computer. Honestly.
Yet again, that is not relevant to the comment I posted. Humans (obviously) do not play "perfect" chess, so computer programs do not need to do so either in order to prevent humans from winning.
You just don't seem to get the relevance of perfect chess:
Humans are striving for perfect chess, and they have improved a lot over the centuries. It's not as if the level of chess playing hasn't improved over the ages.
As long as an opponent doesn't play perfect chess, he's beatable. Even future computers will be beaten by humans (once in a while) as long as they don't play perfect games.
An expert system *taught* (I'm using the term loosely) from the best games of the best players would play like the best masters without the human lapses.
Sigh. An expert system is a Turing type B program. They just don't work for chess because of the complexity involved. Every strong playing chess program we know is type A (brute force) based.
It would be like putting a dictionary program in a spelling bee, and that, I believe, is where we are headed.
I've been trying to tell you for days now that it won't be like that. It's entirely possible to spell perfectly, but it's impossible to play chess perfectly . See the difference?
I write programs because it's my occupation and avocation. Some are simple, and some are sophisticated, but I don't have the hubris to believe I've caused any machine to think.
Try this: mechanical machines reduce the manual labour people have to put up with. Because they reduce our physical workload, we say they do physical work. What kind of human labour do information machines (computers) reduce? Cognitive (mental) labour, IOW thinking. So why isn't it fair to say that computers do actually think, when they reduce the human cognitive workload?
A dogmatic discordian. Now there's a mind-bender.:)
Hehe, I'm not dogmatic at all. Theoretical limits are not dogmas:)
New wave was mostly a commercial reaction to punk rock, and the music industry learned this trick so well most people took grunge for something new while it was just the punk revival the music industry was looking for at that time. Whenever the music cycle devolves into glamour and sweetness, the music industry will pump some "fresh rebel blood" into it.
The music industry reacted in very much the same way to acid house, they treated it like the much expected natural successor to disco and quickly commercialized it into dance, and just as new wave blew the original punk rock out of the charts, dance did the same to the original house/techno.
So predicting popular music is not that hard as long as the music industry can shape our tastes by adapting original gernes to a formula that is easily producable and subsequentially drown us in it. It's the music industry that's shaping these grassroots reactive musical movements into the (predictable) cycles you're talking about.
To restate, yet again, it will take only an incremental improvement in current chess programs before humans can no longer beat them.
As a chess player I can assure you that in general chess players are not impressed when a computer beats them, we would be impressed if a computer played perfect chess or at least wouldn't make obvious strategical mistakes because of horizon effects. Computers are already better at tactics than even the best humans, but they absolutely suck at (long term) strategy; they lack a plan to stick to throughout a game.
So even if computers beat us 99.99% of the time because they can draw us into complex tactical positions; it's mostly the 0.01% that counts, because those are the highly strategical games that hint at perfect chess. (It might very well be that some of those tactical games are perfect too, there is just no way to prove it and these complex tactical victories are not obviously unavoidable.)
If that led you to believe I was arguing about "perfect" chess, I apologize. As I stated, in the same post, I did not understand what your disjointed complaint was.
A perfect game in chess is when you lead your opponent to defeat without ever giving him any chance of escape in the least number of moves possible. I think that's what the confusion is about: in chess "perfect" is defined as absolute perfection, not just being better then everybody/everything else.
As to my disjointed complaint, I'm in the Turing & Kurzweil camp, I believe computers do think (in some very limited way); but that's a philosophical debate, and if we can't have consensus on the definition of "thinking" the debate becomes meaningless.
(And someone who styles himself as the Grand Poobah shouldn't complain about people getting religious.:)
Hey, I'm not complaining, I'm just trying not to let my discordian side take over (but I'm still happy about the conversation getting milder;)
They are your numbers since you are using them to bolster your argument. I believe your numbers are irrelevant since I don't believe there is any need to store all possible positions.
Sigh. You don't need to store all positions, you need to compute them all. Since these computations take more than 10 billion times as long as the universe has existed, it would be very practical to store those positions so you don't have to compute them all over again (during your next move).
The number I give (which is Shannon's estimate, not mine) is widely considered to be the best estimate. My calculations are valid, so if you can't accept them that is your problem, not mine. If you want to attack my estimate, make a better one using published numbers, don't just bullshit around because I don't buy that.
Whoa, Lone Ranger, who is this "we"? You are the one going on and on about perfect games. I'm talking about programs that are *good enough*, and we are nearly there.
You don't believe it's impossible to play perfect chess, read back your own comments (it's the one where you get religious).
You just have a reading comprehension problem, don't you. I don't believe your pessimistic predictions will stand the test of time, nor do I really care how big your numbers are.
Look, it's not my numbers, my part is making an optimistic assumption of hardware capable of computing 10 billion positions per second. With these conservative estimates and optimistic assumptions it's still impossible.
Remember we're talking about playing perfect games, not just being better than the best human. To find a perfect game, you can't just ignore large parts of the search tree and only look a few moves ahead like an ordinary chess program does, you'll have to do an extensive search of each and every move possible before you play your first move, because there can be ways to win playing white or draw playing black which don't look very promising in the beginning.
If you don't understand how something works, you can not duplicate it, although you can produce something that resembles it.
That is besides the point. We don't want to mimic the brain or its biological processes pre ce, we want to duplicate its product, intelligence; we don't have to care about how the human brain produces intelligence (even though it could prove a good starting point).
Modern psychiatry and psychology distinguishes various types of intelligence in humans, so it's fair to try and duplicate only one of those (autonomous) intelligences. It is very well possible to write a program that is capable of learning from its mistakes in playing games (the world champion in checkers is such a program), and it is also possible to write a program that is capable of autonomously learning the rules of various (board) games. If we combined those two concepts, we would have something that duplicates "game intelligence" in humans quite well (except for being awfully slow). Would this be a thinking machine by your standards?
it's your absolutist certitude that amuses me. You aren't the first one to claim that something can't be done and use numbers to back it up. At one time it was known that trains could never travel faster than 60 mph, with all the relevant physics explained.
If you're not confortable with numbers like 10^43 and can't understand that those numbers are incomputable even if we had hardware a hunderd billion times as fast as the hardware available now, that's not my problem; you just don't have the right perspective to appreciate the problem.
And if it takes a very long time to compute positions, so what?
What? Are you serious? 31688800000000000000000000 years? You just haven't got a notion about big numbers, do you?
That does not make the computer intelligent or thinking. It is still a machine that will always follow the same rote instructions without intuition. It is the difference between mimicry and the real thing.
Aha, finally the big word is out. To you, even if a machine "mimics" intelligent behaviour perfectly, it's still a machine, so it can't be intelligent. Do you think human intuition is something special in that it is not determined by rules / laws of nature? What do you think is it that makes the processing in the brain so special a computer can't do it?
Thanks for the information on how Fritz was programmed and for reading Kasparov's mind. Handy talents you have there.
Thank you for appreciating my talents, although I just read the articles on the site linked to in the writeup.
I really doubt you'll see many people interested in gaining chess "insights" from a machine (or in playing chess). It's like gaining insights about how cell K7 relates to cell B4 in a spreadsheet.
Non sequitur. Why do you think people still get on bikes and have races 2000 miles long, considering they know it's allmost impossible to beat someone on a motorbike? Also, why do runners sometimes put weights around their ankles during training? So why exactly shouldn't people still play chess against eachother, and use computers that always beat them during training?
Thank you God. I have always asked You for a message, and now You have answered . . . but I was really hoping for the secret to the universe or a really great stock tip.
Ha ha, I love sarcasm; except what I'm telling is the truth, and the estimate I give about the possible positions in chess is a conservative one (Shannon estimates 10^43 possible positions in a 40 move match). Now consider you have to see every possible position in chess to find the perfect game(s); you can never store all of those positions in any kind of memory (because you would need the whole universe and then some as memory); and it will take a very long time to compute those positions.
Lets pretend our super-giga-hardware can crunch 10 billion positions per second, that would still leave 10^33 seconds to compute all positions, and since there are only about 31556926 seconds in a year, this would make for about 3.16888 * 10^25 years. Our universe has existed for about 1.4 * 10^10 years...
This is why whe know it's impossible to play perfect chess (and no, Moore's law doesn't help us mere mortals much for generations to come).
If you can't explain what thinking is, then you really shouldn't complain if I suggest machines using rule-based programs don't think.
Sigh. The clinical definition of intelligence used to be the "single, general capacity for conceptualization and problem solving" (Gardner, 1983), except Gardener et al. have since then been working on the theory of multiple intelligences, argueing for the existence of several "relatively autonomous" intelligences. So, a) yes, intelligence implicates thinking and intellectual processes, and b) what kind of thinking and intelligence are you exactly talking about (there are at least 8)?
So, you reiterate my point: it's a matter of degree and processor power. You seem to be all bristly and frothy about my comment, but what are you trying to say?
I'm trying to say that you underestimate humans in what they're willing to endure to get better at their sport; that you grossly underestimate the difficulties of chess and overestimate the processor power available to us; and finally that in humans there is no intelligence without thinking, so why doesn't a machine that shows intelligent behaviour think in some way?
Ik know there is heuristics in brute force programs (you can only evaluate the material if you want true brute force), but the point is these programs are still of Turing type A, they do brute force searches. If the programs would be type B they would have a big ruleset and select the appropriate rule(s) to play a given position, but not do brute force searches.
Chess is certainly not an example of AI, it's just a search problem.
If you generalize like this then all problem solving is just searching the problem space. The real problem is that the problem space is too big to do an extensive search.
Minsky at MIT springs to mind as somebody who is trying to nail down this big picture approach to AI but AFAIK nobody has come up with any problem domain that are both complex enough to demonstrate this behaviour but simple enough to understand.
The problem with this is that you're still searching for a particular problem domain; the AI bait 'n switchers will say it's not thinking once you've solved it (because to them it's not thinking if a machine can do it).
The obvious solution would be not to add (more) heuristics to account for these situations but to deal with the horizon effect itself. If you could make the horizon effect go away there would be no need for any heuristics in the game at all.
Of course you can't do that, but you can try to reduce the effect, for instance by switching to a high depth search instantly if you detect that the evaluation isn't changing much during the incremental search. Of course the problem then shifts to not wasting precious time on high depth searches in draw situations.;)
At some point, the decision making (rules) of the best players will be captured in code like other expert applications.
Attempts to make Turing type B (rulebased, heuristic) chess programs have failed so far, all strong playing programs today are of type A (brute force, with perhaps a little heuristics). You can't just make up simple rules for playing chess, those rules will not account for all possible positions on the board; what's good in one position can be instantly losing in another one.
For instance, even the little heuristics in a type A program such as Fritz can be uttlery wrong. Fritz lost the third game to Kasparov because there was a heuristic programmed into it not to push pawns that protect the king, while Kasparov had chosen an opening that resulted in a position where Fritz' only hope in the long term was to do exactly that. Fritz had not enough processor power to look that far into the game, and Kasparov knew it.
After that, what is the point?
The point is to create machines that play chess better than humans do now, so humans can learn from them and get new insights about the game. We know that it's impossible to play perfect chess (because there are more positions possible than there are atoms in the universe), but we also know that what humans (and computers) play today is still very far from perfect.
It won't be the first time, and it won't be a *thinking* machine - just another specialized machine.
Yeah, the classical AI bait 'n switch. First it's really intelligent to play a nice game of chess or do complicated mathematical equations, but as soon as computers can do it it's no longer thinking. The solution would be to sit down and define what thinking exactly is, except nobody has ever succeeded in doing just that.
It does not take a great programmer to write a program that cannot lose at tic-tac-toe against a human.
Yeah, but at the same time no one has written a chess program that cannot lose against a human. Humans still play better chess than computers do, even though the difference between them is getting smaller.
Then I must go back to OS X gnuChess which mocks me every time I play, "You are stupid. I will not play you again."
Try installing Crafty as your chess engine if you're in to masochism. If you really like the pain, install the endgame databases and get those lovely "mate in 57 moves" announcements.
All we have is our one trial, and the test of the theory is really that it provides a plausible mechanism for things turning out the way they did (which would be a tremendous advance compared to the current state of affairs).
This is no different from big-bang and other early universe theories in cosmology. These theories are scientific because they make predictions about the current state of the universe that are falsifiable (CBR intensity and stuff like that), but we can never recreate the original "experiment" for those theories either.
Just like in cosmology, string theory does not only have to provide a plausible explanation for things turning out the way they did, it has to show why this explanation is plausible by being falisifiable, elegance alone simply is not enough.
In seriousness, the length scale at which string theory operates is probably never going to be experimentally accessible.
It doesn't have to be. According to Popper, a theory is scientific if it is falsifiable, it must make "risky" predictions that could turn out to be false. To prove the theory, you try to disprove the risky predictions; if you don't succeed the theory is proven. So you might never be able to positively verify a theory, however it must be falsifiable to be science. It all comes down to that a theory must really predict something with measurable effects to be useful.
Hmm, I hate to rain on your parade, but if I was a voting machine producer trying to rig a vote in your scenario, I would make the machines print out invalid ballots once in a while, electronicly registering that vote to whatever party I support.
Since the vote is anonymous it can't be retracted, and if the US is anything like the Netherlands, it's not that simple to get another ballot form (but then I've been voting electronicly for almost 20 years now, so what do I know:).
Nope, chaos means that the system responds with big changes in its output to very small changes in its input.
If you're a little into math Verhulst' model of biologic growth might help. This model is simply: x(n+1) -> a * x(n) * (1 - x(n)), an iterative model where x(n) is a number between 0 and 1 that indicates the population density at a given step n and a (the Malthusian factor) represents the fertility, a number between 0 and 4.
If you choose a factor a <= 1, the model simulates a dying population, no matter what x(0) you put in, after some iterations it will become 0.
If you pick 1 < a <= 2 the model simulates a stable population, no matter what x(0) you put in, after some iterations in will become 1 - 1/a.
If you pick 2 < a <= 3 the model is still striving for a value of 1 - 1/a but now it will oscillate around this value at an ever smaller absolute distance.
Models with 1 < a <= 3 are balanced, but the interesting stuff starts happening when we pick 3 < a <= 4, because now the model starts behaving chaoticly. If we take a = 3.2 for instance, the model will alternate between the values 0.51304451 and 0.79945549, a lot like the original posters' two alternating states.
Now let's take a = 4 for the sake of argument because the model is then completely chaotic. If we start this with model with x(0) = 0.6875 -> x(12) = 0.925930303 but if we add just 0.0001 x(0) = 0.6876 -> x(12) = 0.5676923. That's a big change in output for a small change in input.
Write a little program and play with this model to really see how randomly it seems to behave, while it's still ruled by a simple deterministic formula.
Maybe they are just words, but I always thought that chaotic and deterministic were opposites.
Not really, chaotic in the mathematical sense means hard to predict, while non-deterministic or random means impossible to predict.
If you're saying that chaos is never truly chaotic, and that it is instead ALWAYS deterministic, then some belief systems (mine actually) will have to be rethought because if there is no such thing as chaos, then there is no such thing as free will.
I'm not saying there is no randomness in the world, I'm only saying that you can't generate true randomness with deterministic systems (like computers) alone, you need a truely random source (like the clicks of a geiger counter) for that.
As for free will, I think Hume's compatibilism could be helpfull to you. Hume very oversimplified defines free will as the freedom to do what one feels like doing (meaning you're still a slave of your passions and feelings, but that's what defines you).
Is free will an illusion or is there really things that are non-deterministic?
The generally accepted interpretation of quantum mechanics claims there is true randomness in the world. However, I personally really don't see how non-determinism would help you in creating a rational definition of free will. If your free will is driven by truely random processes in nature, "rational thought" itself becomes no more than a blind man lead by a fool.
I personally think that (the concept of) "free will" was a nescessary step in our evolution to unify the various unconcious processes in our minds that drive and define us (that generate our feelings, inspirations and insights). It's natures way to assure you that it's really your ideas and feelings, no matter you don't know how exactly they came into being.
The order is the chaotic system itself; chaotic systems are not random but deterministic, their output only appears to be random but is ordered. By wiring the circuits randomly, you do not create randomness in the system, you need (truly) random inputs to do that (and once you do it, the system will no longer fall into or periodicly orbit its attractors like the original poster describes).
Read up about chaos theory and fractal geometry, this is not unusual behaviour in complex systems.
France was in the middle of the turmoil of the Napoleonic era; say what you will about the guy, but he tore down the ancien regime's policy of keeping the peasants ignorant, and set France on the road to democracy.
Acrually it was not Napoleon but the French revolutionaries who guillotuined the ancien regime and its nobility. Right after the revolution the newly formed French republic experiences a lot of internal (anti-revolutionaries) and external (war) pressure, ending in a "reign of terror" by Robespierre's jacobins. After the jacobins are overthrown and Robespierre is executed as a terrorist, a five years period of rebellion and coups follows until Napoleon seizes power in 1799. Napoleon crowns himself emperor of France five years later, declaring "I sealed the gulf of anarchy and unraveled chaos. I purified the revolution and strengthened the monarchy."
While Napoleon did some good for France and Europe, he also caused destruction and killing on a scale not seen before in the world, he took away a lot of the French' newly gained freedoms, and he most certainly did not intend to create a democracy in France.
In short, the great story of the early 19th c., at least in the West, is the "rise of the demos," for the first time in history. And it happened because "a bunch of farmhands," all over the world, wanted to educate themselves.
The French revolution for one started because of new ideas amongst the bourgeoisie who were educating themselves in the growing cities, the peasants were in it mostly because they saw it as the only way to lighten the tax-burden put on them by the ancien regime.
It's very Eastern - the idea that the world is not what you simply perceive.
No, it's not a very eastern idea, it has been around in western philosophy at least since Plato. Plato writes in his Republic of a deeper reality, a world of immutable, eternal forms ("idea" in greek) that can't be directly percieved (book 6, the divided line and the cave allegories). This world of forms is central in platonism and neo-platonism, laying the foundation for the western occult and esoteric schools of thinking.
Realize that Neo empowered smith by destroying him, just as Smith symmetrically empowered Neo by killing him.
This comes also directly from western esoteric philosophy, from hermeticism to be exactly ("As above, so below" on the emerald tablet of Hermes) which teaches that the microcosm and the macrocosm are symmetrically linked (or heaven and earth in the classical magical interpretation). This symmetry is very obvious between the architect and the oracle also. Hermeticism also teaches enlightment, by the way; as does gnosticism which is also to be found in all three episodes of the matrix.
There are buddhist influences in the matrix, especially in revolutions, but it's not as much as one would think. The Wachowskis like putting (and sometimes hiding) references to philosophy in the matrix; don't underestimate what wonderful weirdness there is to be found in the western schools they most often bring up.
Substantially different numbers are published (I found 75,000, 140,000 and 200,000), I guess demographics wasn't high on the priority list in 1600, and these are all estimates. To be on the safe side I picked the lowest guess I found.:)
Uhh, mr. Fawkes had about 2500kg of gunpowder stacked up (which the article says is equivalent to 2500kg TNT), I don't see anyone carrying that into a Starbucks easily.
London might have had a population of only 75000 people in 1605, but an explosion with a blast radius of 1/3 of a mile and a damage radius of 2/3 of a mile right in the centre of the city would have probably killed and wounded more people than the WTC attack did.
Having the MSH- or any other shell, by MS or someone else- return an object doesn't hinder the ability for a user to read the output. See, returning an object gives the end-user using the shell as well as developers a lot more power.... But, in the case you did a pipe, you wouldn't have to get primitive and parse that text. Why write code that renders output into text you just have to go an un-render?
You are missing the point. The second reason for having plaintext protocols an output (the first being of course human readability) is because it gives more expressive power. Parsing plaintext is not very expensive, and using plaintext protocols means you do not have to worry so much about a representation that's acceptable to any program that might use it. Thirdly, regarding the future, it's usually a lot easier to extend a plaintext protocol than it is to extend a binary protocol.
You might think these are silly arguments, however these are generally the reasons for why the large majority of internet protocols are plaintext protocols.
There would be no need to write a script that casts/converts what was returned by one method/program to something another could understand, unless you had actual processing to do, which sometimes you will... But that is sometimes the case no matter if you're piping a string or a real object. In fact, with something like the setup in MSH, you'll have to do less fiddling to get this.
Ok, suppose our boss comes in and we have to delete all evidence (meaning all files that are not c files) from our homedirs. I go cd && ls | grep -v '.c$\|.h$' | xargs rm -f, what do you do? Because I'm working in plaintext, I've got a set of powerful and versatile i/o manipulation tools available. I can parse anything once I've mastered a tool like grep or sed, so I can quickly pipe stuff together no one ever dreamed of (especially with xargs). You need a lot more information (typenames, methods, etc) to manipulate objects, and you'll probably produce a much longer oneliner than I did because you'll be writing a foreach loop with filename parsing inside.
Much to my annoyance, if I pipe an "ls" in psh, I still have to do parsing and such. I'd much rather me able to do-
ls | foreach $file (@files) { doSomething($file); }
It's not much different from ls | xargs doSomething if doSomething takes multiple arguments, or ls | while read file; do doSomething $file; done if nothing else helps;)
A big, cleverly written middle man library isn't neccesary. Yes, it would be for C++ and probably Java, but not for.NET.
Wouldn't people write scripts to automate things like my example above in MSH (delete all files that don't have an extension that is present within a given set)? Wouldn't there be whole libraries of these if there is no unified and staightforward way to filter and process i/o linearly (in a pipe, not in a foreach loop)?
By the way, "Bob" exiled Eris to the bog of eternal stench a long time ago. The X-ists have lost- forever!
If you're a bit into this stuff you might notice a carry ranks with all of the opposing parties. This is what Eris truly demands of us, not taking one side, and of course not being neutral. About the SubGenii: every party in this neverending comedy has it's UFO-cults, even the Illuminati have some weird cousins in fringe freemansonry who sometimes even kill themselves when a comet passes by. All this assures us that Erisian randomness rules supreme and Eris still laughs at us from her shiny gold spaceship.
A,NET object is just an object. Really not all that much more overhead compared to using printfs or spitting it out to a file and then writing your own code to process that text. Why not cut out the middle man?
For one, the middle man enables you -the shell user- to just type a command, perhaps type some more input, and read the output.
Secondly, the middle man enables you to easily chain (pipe the i/o of) various tools/commands in ways the authors of these tools might not have expected. If tool A creates a.NET object of type X as output but tool B expects a.NET object of type Y as input, there is no easy way of chaining these tools on the command line by putting some prefab filter between them, you'll have to write a more or less elaborate (middle man) script.
By creating an interactive layer over.NET, what MS has really done is bringing an interactive scripting language like basic or python into the world. This "shell" will be perfect for scripting, but be rather not so good for interactive use if MS doesn't provide a big, cleverly written middle man library. If MS doesn't provide a good middle man library, the shell will be pretty useless without a custom library, probably creating very diverse shell environments (a lot like the situation with scripted IRC clients).
Slashdot Mirror
Hehe. Of course chess is mental, but that doesn't take away that just as you can blunder by fumbling a ball in baseball, you can blunder by fumbling a position in chess. There are hundreds of examples of (grand)masters losing a game to a much lower ranking opponent. In chess there is the added stress of not only playing against an opponent but playing against a clock also, and it's the mental stress that's mostly responsible for the mistakes you make, not only in chess but in a wide range of physical sports also.
Some? Crossword puzzles maybe. When I run into problems, I usually try to remember similar situations and the various outcomes (or at least I think I do).
That would be a concious process, but the crosswords you mention is perhaps (interestingly) related to where I mostly detect these unconcious processes in myself, namely when I'm trying to remember words (in a foreign language) and catch myself mumbling things like: malag maleg malig ahh, it's malignant. I'm unconciously doing a brute force on the phonemes of a word until it "sounds" (feels) right. There are other examples, but I've seen other people do this too, so maybe you're familliar with it also.
The rules of warfare are myriad and changeable, and the possible environments nearly infinite. The rules of chess are fixed and the battlefield unchangeable.
Agreed, but we've already discussed the number of possible positions in a normal game of chess. Isn't it fair to call these a myrad of possibilities also? ;)
But all that still does not address your claim that anything that reduces a human's cognitive workload is capable of thinking.
Let's simply define thinking as excercizing your intelligence(s), IOW making use of your abilities to abstract, solve problems, etc. Any machine that reduces the need for a human to perform these tasks while the abstract models are still being produced and/or the problems are still being solved, must be performing (part of) these tasks for the human; because performing these tasks is defined as thinking, the machine must be thinking to a certain extent.
Well, I'm sure we've broken my previous record for ongoing arguments on Slashdot, and it seems we've agreed to disagree. I'd wish you a happy Thanksgiving, but your spelling of "labour" suggests you're a Brit, and the holiday is only for us colonists. :)
I'm much better at starting discussions than at ending them, and even though I tried to keep this last post short, look what happened. I'm not a Brit but a Dutchie by the way; it was them Brits who gave us trouble in Nieuw Amsterdam and rebaptized it to New York, so don't get me started about them :) Pleasant holidays.
Mallory can configure her LDAP server to not only give her root on you box, but also to remotely mount filesystems on your box. Mallory mounts her trojans over your bin directory and waits for you to start one, or also mounts a root crontab that starts a trojan automaticly. No your box IS running services, and Mallory owns it.
Ok, I'll try to explain it one more time. In chess (like in most sports) there is absolutely no guarantee that the better player (the higher ranking player) will always win. Even worse, you can't even prove that looking ahead further in the search tree produces better moves (because of horizon effects, the move you think is good might prove very bad if you look ahead one move further, ad nauseam). The only way of being absolutely unbeatable is to play perfect chess, IOW to exhaust the search tree completely.
I would think that someone who believes computers can think would see the waste in brute force solutions and be able to imagine the future of expert systems.
I don't see it as waste, in fact I think some of the thinking we humans do is of brute-force type, we just don't normally realize it because they are subconcious processes. The problem with rule-based systems is emergent properties; if you combine some individually simple rules, complex situations can arrise. That's why expert systems don't work for problems like chess: it's like feeding a general with knowledge of famous battles of the past and expecting him to win every future battle.
In earlier years, I built a calculator and a computer from scratch using wire-wrap, so I can attest that a calculator is nothing more than an electronic abacus or slide rule, and a computer is nothing more than a fancy calculator.
Yes, but very similarly, if you disect/study a living human brain you will only find braincells but no thought. If a brain is only neurons, how can it think? My answer: intelligence is an emergent property (also).
And as a chess player that really played competition and not just recreatively, I can say a lot. There are a lot of people who practice chess every day and study it two hours a day or more.
I can understand your point (though I don't agree completely) that it's not a lot of fun to recreatively play chess against a much stronger opponent. But if you play chess competatively and want to improve your tactics, there is nothing like playing a lot of games against a strong computer. Honestly.
Yet again, that is not relevant to the comment I posted. Humans (obviously) do not play "perfect" chess, so computer programs do not need to do so either in order to prevent humans from winning.
You just don't seem to get the relevance of perfect chess:
Humans are striving for perfect chess, and they have improved a lot over the centuries. It's not as if the level of chess playing hasn't improved over the ages.
As long as an opponent doesn't play perfect chess, he's beatable. Even future computers will be beaten by humans (once in a while) as long as they don't play perfect games.
An expert system *taught* (I'm using the term loosely) from the best games of the best players would play like the best masters without the human lapses.
Sigh. An expert system is a Turing type B program. They just don't work for chess because of the complexity involved. Every strong playing chess program we know is type A (brute force) based.
It would be like putting a dictionary program in a spelling bee, and that, I believe, is where we are headed.
I've been trying to tell you for days now that it won't be like that. It's entirely possible to spell perfectly, but it's impossible to play chess perfectly . See the difference?
I write programs because it's my occupation and avocation. Some are simple, and some are sophisticated, but I don't have the hubris to believe I've caused any machine to think.
Try this: mechanical machines reduce the manual labour people have to put up with. Because they reduce our physical workload, we say they do physical work. What kind of human labour do information machines (computers) reduce? Cognitive (mental) labour, IOW thinking. So why isn't it fair to say that computers do actually think, when they reduce the human cognitive workload?
A dogmatic discordian. Now there's a mind-bender. :)
Hehe, I'm not dogmatic at all. Theoretical limits are not dogmas :)
The music industry reacted in very much the same way to acid house, they treated it like the much expected natural successor to disco and quickly commercialized it into dance, and just as new wave blew the original punk rock out of the charts, dance did the same to the original house/techno.
So predicting popular music is not that hard as long as the music industry can shape our tastes by adapting original gernes to a formula that is easily producable and subsequentially drown us in it. It's the music industry that's shaping these grassroots reactive musical movements into the (predictable) cycles you're talking about.
As a chess player I can assure you that in general chess players are not impressed when a computer beats them, we would be impressed if a computer played perfect chess or at least wouldn't make obvious strategical mistakes because of horizon effects. Computers are already better at tactics than even the best humans, but they absolutely suck at (long term) strategy; they lack a plan to stick to throughout a game.
So even if computers beat us 99.99% of the time because they can draw us into complex tactical positions; it's mostly the 0.01% that counts, because those are the highly strategical games that hint at perfect chess. (It might very well be that some of those tactical games are perfect too, there is just no way to prove it and these complex tactical victories are not obviously unavoidable.)
If that led you to believe I was arguing about "perfect" chess, I apologize. As I stated, in the same post, I did not understand what your disjointed complaint was.
A perfect game in chess is when you lead your opponent to defeat without ever giving him any chance of escape in the least number of moves possible. I think that's what the confusion is about: in chess "perfect" is defined as absolute perfection, not just being better then everybody/everything else.
As to my disjointed complaint, I'm in the Turing & Kurzweil camp, I believe computers do think (in some very limited way); but that's a philosophical debate, and if we can't have consensus on the definition of "thinking" the debate becomes meaningless.
(And someone who styles himself as the Grand Poobah shouldn't complain about people getting religious. :)
Hey, I'm not complaining, I'm just trying not to let my discordian side take over (but I'm still happy about the conversation getting milder ;)
Sigh. You don't need to store all positions, you need to compute them all. Since these computations take more than 10 billion times as long as the universe has existed, it would be very practical to store those positions so you don't have to compute them all over again (during your next move).
The number I give (which is Shannon's estimate, not mine) is widely considered to be the best estimate. My calculations are valid, so if you can't accept them that is your problem, not mine. If you want to attack my estimate, make a better one using published numbers, don't just bullshit around because I don't buy that.
Whoa, Lone Ranger, who is this "we"? You are the one going on and on about perfect games. I'm talking about programs that are *good enough*, and we are nearly there.
You don't believe it's impossible to play perfect chess, read back your own comments (it's the one where you get religious).
Look, it's not my numbers, my part is making an optimistic assumption of hardware capable of computing 10 billion positions per second. With these conservative estimates and optimistic assumptions it's still impossible.
Remember we're talking about playing perfect games, not just being better than the best human. To find a perfect game, you can't just ignore large parts of the search tree and only look a few moves ahead like an ordinary chess program does, you'll have to do an extensive search of each and every move possible before you play your first move, because there can be ways to win playing white or draw playing black which don't look very promising in the beginning.
If you don't understand how something works, you can not duplicate it, although you can produce something that resembles it.
That is besides the point. We don't want to mimic the brain or its biological processes pre ce, we want to duplicate its product, intelligence; we don't have to care about how the human brain produces intelligence (even though it could prove a good starting point).
Modern psychiatry and psychology distinguishes various types of intelligence in humans, so it's fair to try and duplicate only one of those (autonomous) intelligences. It is very well possible to write a program that is capable of learning from its mistakes in playing games (the world champion in checkers is such a program), and it is also possible to write a program that is capable of autonomously learning the rules of various (board) games. If we combined those two concepts, we would have something that duplicates "game intelligence" in humans quite well (except for being awfully slow). Would this be a thinking machine by your standards?
If you're not confortable with numbers like 10^43 and can't understand that those numbers are incomputable even if we had hardware a hunderd billion times as fast as the hardware available now, that's not my problem; you just don't have the right perspective to appreciate the problem.
And if it takes a very long time to compute positions, so what?
What? Are you serious? 31688800000000000000000000 years? You just haven't got a notion about big numbers, do you?
That does not make the computer intelligent or thinking. It is still a machine that will always follow the same rote instructions without intuition. It is the difference between mimicry and the real thing.
Aha, finally the big word is out. To you, even if a machine "mimics" intelligent behaviour perfectly, it's still a machine, so it can't be intelligent. Do you think human intuition is something special in that it is not determined by rules / laws of nature? What do you think is it that makes the processing in the brain so special a computer can't do it?
I really doubt you'll see many people interested in gaining chess "insights" from a machine (or in playing chess). It's like gaining insights about how cell K7 relates to cell B4 in a spreadsheet.
Non sequitur. Why do you think people still get on bikes and have races 2000 miles long, considering they know it's allmost impossible to beat someone on a motorbike? Also, why do runners sometimes put weights around their ankles during training? So why exactly shouldn't people still play chess against eachother, and use computers that always beat them during training?
Thank you God. I have always asked You for a message, and now You have answered . . . but I was really hoping for the secret to the universe or a really great stock tip.
Ha ha, I love sarcasm; except what I'm telling is the truth, and the estimate I give about the possible positions in chess is a conservative one (Shannon estimates 10^43 possible positions in a 40 move match). Now consider you have to see every possible position in chess to find the perfect game(s); you can never store all of those positions in any kind of memory (because you would need the whole universe and then some as memory); and it will take a very long time to compute those positions.
Lets pretend our super-giga-hardware can crunch 10 billion positions per second, that would still leave 10^33 seconds to compute all positions, and since there are only about 31556926 seconds in a year, this would make for about 3.16888 * 10^25 years. Our universe has existed for about 1.4 * 10^10 years...
This is why whe know it's impossible to play perfect chess (and no, Moore's law doesn't help us mere mortals much for generations to come).
If you can't explain what thinking is, then you really shouldn't complain if I suggest machines using rule-based programs don't think.
Sigh. The clinical definition of intelligence used to be the "single, general capacity for conceptualization and problem solving" (Gardner, 1983), except Gardener et al. have since then been working on the theory of multiple intelligences, argueing for the existence of several "relatively autonomous" intelligences. So, a) yes, intelligence implicates thinking and intellectual processes, and b) what kind of thinking and intelligence are you exactly talking about (there are at least 8)?
So, you reiterate my point: it's a matter of degree and processor power. You seem to be all bristly and frothy about my comment, but what are you trying to say?
I'm trying to say that you underestimate humans in what they're willing to endure to get better at their sport; that you grossly underestimate the difficulties of chess and overestimate the processor power available to us; and finally that in humans there is no intelligence without thinking, so why doesn't a machine that shows intelligent behaviour think in some way?
Chess is certainly not an example of AI, it's just a search problem.
If you generalize like this then all problem solving is just searching the problem space. The real problem is that the problem space is too big to do an extensive search.
Minsky at MIT springs to mind as somebody who is trying to nail down this big picture approach to AI but AFAIK nobody has come up with any problem domain that are both complex enough to demonstrate this behaviour but simple enough to understand.
The problem with this is that you're still searching for a particular problem domain; the AI bait 'n switchers will say it's not thinking once you've solved it (because to them it's not thinking if a machine can do it).
Of course you can't do that, but you can try to reduce the effect, for instance by switching to a high depth search instantly if you detect that the evaluation isn't changing much during the incremental search. Of course the problem then shifts to not wasting precious time on high depth searches in draw situations. ;)
Attempts to make Turing type B (rulebased, heuristic) chess programs have failed so far, all strong playing programs today are of type A (brute force, with perhaps a little heuristics). You can't just make up simple rules for playing chess, those rules will not account for all possible positions on the board; what's good in one position can be instantly losing in another one.
For instance, even the little heuristics in a type A program such as Fritz can be uttlery wrong. Fritz lost the third game to Kasparov because there was a heuristic programmed into it not to push pawns that protect the king, while Kasparov had chosen an opening that resulted in a position where Fritz' only hope in the long term was to do exactly that. Fritz had not enough processor power to look that far into the game, and Kasparov knew it.
After that, what is the point?
The point is to create machines that play chess better than humans do now, so humans can learn from them and get new insights about the game. We know that it's impossible to play perfect chess (because there are more positions possible than there are atoms in the universe), but we also know that what humans (and computers) play today is still very far from perfect.
It won't be the first time, and it won't be a *thinking* machine - just another specialized machine.
Yeah, the classical AI bait 'n switch. First it's really intelligent to play a nice game of chess or do complicated mathematical equations, but as soon as computers can do it it's no longer thinking. The solution would be to sit down and define what thinking exactly is, except nobody has ever succeeded in doing just that.
It does not take a great programmer to write a program that cannot lose at tic-tac-toe against a human.
Yeah, but at the same time no one has written a chess program that cannot lose against a human. Humans still play better chess than computers do, even though the difference between them is getting smaller.
Try installing Crafty as your chess engine if you're in to masochism. If you really like the pain, install the endgame databases and get those lovely "mate in 57 moves" announcements.
This is no different from big-bang and other early universe theories in cosmology. These theories are scientific because they make predictions about the current state of the universe that are falsifiable (CBR intensity and stuff like that), but we can never recreate the original "experiment" for those theories either.
Just like in cosmology, string theory does not only have to provide a plausible explanation for things turning out the way they did, it has to show why this explanation is plausible by being falisifiable, elegance alone simply is not enough.
It doesn't have to be. According to Popper, a theory is scientific if it is falsifiable, it must make "risky" predictions that could turn out to be false. To prove the theory, you try to disprove the risky predictions; if you don't succeed the theory is proven. So you might never be able to positively verify a theory, however it must be falsifiable to be science. It all comes down to that a theory must really predict something with measurable effects to be useful.
Since the vote is anonymous it can't be retracted, and if the US is anything like the Netherlands, it's not that simple to get another ballot form (but then I've been voting electronicly for almost 20 years now, so what do I know :).
Nope, chaos means that the system responds with big changes in its output to very small changes in its input.
If you're a little into math Verhulst' model of biologic growth might help. This model is simply: x(n+1) -> a * x(n) * (1 - x(n)), an iterative model where x(n) is a number between 0 and 1 that indicates the population density at a given step n and a (the Malthusian factor) represents the fertility, a number between 0 and 4.
If you choose a factor a <= 1, the model simulates a dying population, no matter what x(0) you put in, after some iterations it will become 0.
If you pick 1 < a <= 2 the model simulates a stable population, no matter what x(0) you put in, after some iterations in will become 1 - 1/a.
If you pick 2 < a <= 3 the model is still striving for a value of 1 - 1/a but now it will oscillate around this value at an ever smaller absolute distance.
Models with 1 < a <= 3 are balanced, but the interesting stuff starts happening when we pick 3 < a <= 4, because now the model starts behaving chaoticly. If we take a = 3.2 for instance, the model will alternate between the values 0.51304451 and 0.79945549, a lot like the original posters' two alternating states.
Now let's take a = 4 for the sake of argument because the model is then completely chaotic. If we start this with model with x(0) = 0.6875 -> x(12) = 0.925930303 but if we add just 0.0001 x(0) = 0.6876 -> x(12) = 0.5676923. That's a big change in output for a small change in input.
Write a little program and play with this model to really see how randomly it seems to behave, while it's still ruled by a simple deterministic formula.
Maybe they are just words, but I always thought that chaotic and deterministic were opposites.
Not really, chaotic in the mathematical sense means hard to predict, while non-deterministic or random means impossible to predict.
If you're saying that chaos is never truly chaotic, and that it is instead ALWAYS deterministic, then some belief systems (mine actually) will have to be rethought because if there is no such thing as chaos, then there is no such thing as free will.
I'm not saying there is no randomness in the world, I'm only saying that you can't generate true randomness with deterministic systems (like computers) alone, you need a truely random source (like the clicks of a geiger counter) for that.
As for free will, I think Hume's compatibilism could be helpfull to you. Hume very oversimplified defines free will as the freedom to do what one feels like doing (meaning you're still a slave of your passions and feelings, but that's what defines you).
Is free will an illusion or is there really things that are non-deterministic?
The generally accepted interpretation of quantum mechanics claims there is true randomness in the world. However, I personally really don't see how non-determinism would help you in creating a rational definition of free will. If your free will is driven by truely random processes in nature, "rational thought" itself becomes no more than a blind man lead by a fool.
I personally think that (the concept of) "free will" was a nescessary step in our evolution to unify the various unconcious processes in our minds that drive and define us (that generate our feelings, inspirations and insights). It's natures way to assure you that it's really your ideas and feelings, no matter you don't know how exactly they came into being.
Read up about chaos theory and fractal geometry, this is not unusual behaviour in complex systems.
Acrually it was not Napoleon but the French revolutionaries who guillotuined the ancien regime and its nobility. Right after the revolution the newly formed French republic experiences a lot of internal (anti-revolutionaries) and external (war) pressure, ending in a "reign of terror" by Robespierre's jacobins. After the jacobins are overthrown and Robespierre is executed as a terrorist, a five years period of rebellion and coups follows until Napoleon seizes power in 1799. Napoleon crowns himself emperor of France five years later, declaring "I sealed the gulf of anarchy and unraveled chaos. I purified the revolution and strengthened the monarchy."
While Napoleon did some good for France and Europe, he also caused destruction and killing on a scale not seen before in the world, he took away a lot of the French' newly gained freedoms, and he most certainly did not intend to create a democracy in France.
In short, the great story of the early 19th c., at least in the West, is the "rise of the demos," for the first time in history. And it happened because "a bunch of farmhands," all over the world, wanted to educate themselves.
The French revolution for one started because of new ideas amongst the bourgeoisie who were educating themselves in the growing cities, the peasants were in it mostly because they saw it as the only way to lighten the tax-burden put on them by the ancien regime.
(And no, I'm not French :)
No, it's not a very eastern idea, it has been around in western philosophy at least since Plato. Plato writes in his Republic of a deeper reality, a world of immutable, eternal forms ("idea" in greek) that can't be directly percieved (book 6, the divided line and the cave allegories). This world of forms is central in platonism and neo-platonism, laying the foundation for the western occult and esoteric schools of thinking.
Realize that Neo empowered smith by destroying him, just as Smith symmetrically empowered Neo by killing him.
This comes also directly from western esoteric philosophy, from hermeticism to be exactly ("As above, so below" on the emerald tablet of Hermes) which teaches that the microcosm and the macrocosm are symmetrically linked (or heaven and earth in the classical magical interpretation). This symmetry is very obvious between the architect and the oracle also. Hermeticism also teaches enlightment, by the way; as does gnosticism which is also to be found in all three episodes of the matrix.
There are buddhist influences in the matrix, especially in revolutions, but it's not as much as one would think. The Wachowskis like putting (and sometimes hiding) references to philosophy in the matrix; don't underestimate what wonderful weirdness there is to be found in the western schools they most often bring up.
Substantially different numbers are published (I found 75,000, 140,000 and 200,000), I guess demographics wasn't high on the priority list in 1600, and these are all estimates. To be on the safe side I picked the lowest guess I found. :)
London might have had a population of only 75000 people in 1605, but an explosion with a blast radius of 1/3 of a mile and a damage radius of 2/3 of a mile right in the centre of the city would have probably killed and wounded more people than the WTC attack did.
You are missing the point. The second reason for having plaintext protocols an output (the first being of course human readability) is because it gives more expressive power. Parsing plaintext is not very expensive, and using plaintext protocols means you do not have to worry so much about a representation that's acceptable to any program that might use it. Thirdly, regarding the future, it's usually a lot easier to extend a plaintext protocol than it is to extend a binary protocol.
You might think these are silly arguments, however these are generally the reasons for why the large majority of internet protocols are plaintext protocols.
There would be no need to write a script that casts/converts what was returned by one method/program to something another could understand, unless you had actual processing to do, which sometimes you will... But that is sometimes the case no matter if you're piping a string or a real object. In fact, with something like the setup in MSH, you'll have to do less fiddling to get this.
Ok, suppose our boss comes in and we have to delete all evidence (meaning all files that are not c files) from our homedirs. I go cd && ls | grep -v '.c$\|.h$' | xargs rm -f, what do you do? Because I'm working in plaintext, I've got a set of powerful and versatile i/o manipulation tools available. I can parse anything once I've mastered a tool like grep or sed, so I can quickly pipe stuff together no one ever dreamed of (especially with xargs). You need a lot more information (typenames, methods, etc) to manipulate objects, and you'll probably produce a much longer oneliner than I did because you'll be writing a foreach loop with filename parsing inside.
Much to my annoyance, if I pipe an "ls" in psh, I still have to do parsing and such. I'd much rather me able to do-
ls | foreach $file (@files) { doSomething($file); }
It's not much different from ls | xargs doSomething if doSomething takes multiple arguments, or ls | while read file; do doSomething $file; done if nothing else helps ;)
A big, cleverly written middle man library isn't neccesary. Yes, it would be for C++ and probably Java, but not for .NET.
Wouldn't people write scripts to automate things like my example above in MSH (delete all files that don't have an extension that is present within a given set)? Wouldn't there be whole libraries of these if there is no unified and staightforward way to filter and process i/o linearly (in a pipe, not in a foreach loop)?
By the way, "Bob" exiled Eris to the bog of eternal stench a long time ago. The X-ists have lost- forever!
If you're a bit into this stuff you might notice a carry ranks with all of the opposing parties. This is what Eris truly demands of us, not taking one side, and of course not being neutral. About the SubGenii: every party in this neverending comedy has it's UFO-cults, even the Illuminati have some weird cousins in fringe freemansonry who sometimes even kill themselves when a comet passes by. All this assures us that Erisian randomness rules supreme and Eris still laughs at us from her shiny gold spaceship.
For one, the middle man enables you -the shell user- to just type a command, perhaps type some more input, and read the output.
Secondly, the middle man enables you to easily chain (pipe the i/o of) various tools/commands in ways the authors of these tools might not have expected. If tool A creates a .NET object of type X as output but tool B expects a .NET object of type Y as input, there is no easy way of chaining these tools on the command line by putting some prefab filter between them, you'll have to write a more or less elaborate (middle man) script.
By creating an interactive layer over .NET, what MS has really done is bringing an interactive scripting language like basic or python into the world. This "shell" will be perfect for scripting, but be rather not so good for interactive use if MS doesn't provide a big, cleverly written middle man library. If MS doesn't provide a good middle man library, the shell will be pretty useless without a custom library, probably creating very diverse shell environments (a lot like the situation with scripted IRC clients).